Deep Viscosity

1-5 minutesPaperCodeNeed >5GB

Viscosity prediction for antibodies

Readiness

Readiness: Need >5GB
Smoke: passed
Artifacts: passed
Production: need_gt5gb_keras2_anarci_anarcii_runtime_pending
Storage: 5-20GB
Strategy: large shared TensorFlow/Keras 2.x + ANARCI/ANARCII SIF/overlay
Next: Containerize a clean Keras 2.x plus ANARCI/ANARCII runtime, run Deep Viscosity end-to-end through Slurm until final class CSV is produced, then wire runner/API.

SOP agent cloned upstream DeepViscosity revision bf7a500da445b93c7f2546a3fa62062ac10e4fc0 at /media/nik/seagate_nik/bio_server/tools/deep-viscosity. Source about 108M; tool-local py39 deps about 287M. A py312 ANARCII test venv reached about 6.5G because ANARCII pulled CUDA PyTorch and was not validated. Best model-load env found: /home/nik/micromamba-root/envs/canya-py39 with TensorFlow 2.10.1/Keras 2.10.0 plus deps_py39 for Biopython/dill; legacy ANARCI at /home/nik/bio_server_venvs/anarci/bin/ANARCI. Direct predictor was attempted and produced partial aligned CSVs and DeepSP_descriptors.csv under /media/nik/seagate_nik/bio_server/tools/deep-viscosity/artifacts/validation_20260701_codex_direct, but final DeepViscosity_classes.csv was not validated. No Slurm job submitted. Modern TF/Keras 3 cannot deserialize upstream Keras 2.11 JSON models. 2026-07-07 17:16 +0800: SOP Runner Adapter/Blocked pass: moved from Production Blocked to Need >5GB because ANARCII runtime staging already reached about 6.5GB and final DeepViscosity_classes.csv was not validated through Slurm.